Wireless network using neural network for channel state feedback

Abstract

To provide a method and a device to implement a channel estimate process that is adaptive to a change in a state, has high efficiency, and is robust.SOLUTION: A wireless system (100) uses deep neural networks (DNN) (122, 128) to provide channel state information (CSI) estimate feedback between a base station (BS) (108) and UE (110). The UE determines a CSI estimate (134) from CSI pilot signaling from the BS. The CSI estimate is processed by the DNNs of the UE, and CSF output (136) representing one or more predicted future CSI estimates is generated and is wirelessly transmitted to the BS. Subsequently, one or more DNNs of the BS process the received CSF output, generate one or more recovered predicted future CSI estimates (138), and use them to control one or more MIMO processes in the BS.SELECTED DRAWING: Figure 1

Description

[Background technology]

[0001] background Wireless communication systems often encounter several frequency-dependent signal propagation challenges, such as path loss, scattering, signal diffraction, penetration loss, etc. Many radio transmission schemes that rely on high frequencies, such as those compliant with the Third Generation Partnership (3GPP®) Long Term Evolution (LTE) and Fifth Generation New Radio (5G NR) cellular standards, or certain International Electronics and Electrical Engineers (IEEE) 802.11 wireless local area network (WLAN) standards, use various multiple-input multiple-output (MIMO) techniques to mitigate such signal propagation challenges.

[0002] Many MIMO technologies, such as beamforming, space-time coding, and multi-user MIMO (MU-MIMO) technologies, rely on, or at least benefit from, understanding or characterizing how wireless signals propagate in the current signal propagation environment at one or more corresponding carrier frequencies of an identified channel. Typically, this channel estimate is provided as channel state information (CSI). The CSI often takes the form of one or more matrices, with each matrix entry storing information representing the transfer function, or more specifically, the channel frequency response (CFR), of a corresponding carrier frequency. To determine the CSI of a channel, a transmitter wirelessly transmits one or more CSI pilot symbols, such as long training symbols (LTFs) in IEEE 802.11-based systems, to a receiver, and then uses the received form of the transmitted CSI pilot symbols to calculate at least one CSI estimate for the corresponding carrier frequency used to transmit the one or more CSI pilot symbols. The receiver can utilize the CSI estimate to manage the MIMO reception process for a particular channel. This CSI estimate may be provided wirelessly to the transmitting device, allowing the transmitting device to manage one or more MIMO transmission processes accordingly (this feedback process is typically referred to as "channel state feedback" (CSF)).

[0003] The entire process of transmitting pilot symbols, calculating CSI estimates from received pilot symbols, and reporting the CSI estimates to a transmitter is typically implemented via a modular approach, where each algorithm, step, or stage of the process, is individually "handcrafted" by one or more designers. The relative complexity of each step typically leads to corresponding complexity in designing, testing, and implementing a hard-coded implementation of the process. Furthermore, complex algorithmic computation of CSI estimates at the receiver can consume significant resources at the receiver, while frequent over-the-air transmission of CSI estimates in a generic format can consume significant bandwidth on the channel connecting the receiver to the transmitter. Therefore, it can be challenging to design and implement a robust channel estimation process that is highly adaptive to changing conditions and efficient for reducing transmission latency and resource consumption. Summary of the Invention

[0004] Overview of the embodiment According to some embodiments, a computer-implemented method in a first device includes receiving an indication of a neural network architecture configuration in response to providing capability information representing at least one capability of the first device to an infrastructure component; implementing the neural network architecture configuration in a transmit neural network of the first device; receiving a representation of a channel state information (CSI) estimate as an input to the transmit neural network; and implementing the transmit neural network architecture configuration in the transmit neural network. generating a first output based on a representation of the CSI estimate, the first output representing a condensed version of a representation of a prediction of the CSI estimate for a future time point; and controlling a radio frequency (RF) antenna interface of the first device to transmit a first RF signal representing the first output for reception by the second device.

[0005] In various embodiments, the method may further include one or more of the following aspects: The method may further include algorithmically determining a CSI estimate based on one or more RF signals received from the second device; The first output further represents a prediction of the CSI estimate at a future time point (i.e., a predicted future CSI estimate); Generating the first output further includes generating the first output at the transmit neural network further based on a representation of a scheduling latency of a multiple-input multiple-output (MIMO) process of the second device provided as an input to the transmit neural network; The transmit neural network receives the representation of the scheduling latency as an input; A neural network architecture configuration is selected for the transmit neural network from a plurality of candidate neural network architecture configurations based on the scheduling latency; The at least one capability represented by the capability information includes at least one of antenna capability, processing capability, power capability, or sensor capability; The neural network architecture configuration is selected from a plurality of neural network architecture configurations based on at least one of the following: at least one capability of the first device, at least one capability of the second device, the frequency or band of the channel represented by the CSI estimate, or a current signal propagation environment of the first device. Receiving an indication of a neural network architecture configuration includes at least one of receiving an identifier associated with one of a plurality of candidate neural network architecture configurations stored locally on the first device, or receiving data representing parameters of the neural network architecture configuration. Generating a first output includes generating the first output in the transmit neural network further based on at least one of the following: sensor data input to the transmit neural network from one or more sensors of the first device, or at least one current operating parameter of the RF antenna interface.The method further includes participating in joint training with a transmit neural network architecture configuration of the transmit neural network and a receive neural network architecture configuration of the receive neural network of the second device. The method also includes receiving a representation of a CSI pilot signal as an input to the receive neural network of the first device, and generating a second output at the receive neural network based on the representation of the CSI pilot signal, the second output including a representation of a CSI estimate. The generating a second output further includes generating the second output at the receive neural network based on at least one of: sensor data from one or more sensors of the first device, a carrier frequency of a channel associated with the CSI estimate, or an operating parameter of an antenna interface of the first device. The transmit neural network is a deep neural network (DNN).

[0006] According to some embodiments, a computer-implemented method in a first device includes receiving, in response to providing capability information representing at least one capability of the first device to an infrastructure component, an indication of a neural network architecture configuration; implementing the neural network architecture configuration in a receive neural network of the first device; receiving, at a radio frequency (RF) antenna interface of the first device, a first RF signal representing a condensed representation of a predicted future channel state information (CSI) estimate from a second device; providing the representation of the first RF signal as an input to the receive neural network; and implementing, at the receive neural network, the representation of the first RF signal representing a condensed representation of a predicted future channel state information (CSI) estimate. generating a predicted future CSI estimate based on input to the neural network; and managing at least one multiple-input multiple-output (MIMO) process at the first device based on the predicted future CSI estimate.

[0007] In various embodiments, the method may further include one or more of the following aspects: the neural network architecture configuration is selected from a plurality of neural network architecture configurations based on at least one of the following: at least one capability of the first device, at least one capability of the second device, a channel frequency or band represented by the predicted future CSI estimate, or a current signal propagation environment of the first device; receiving an indication of a neural network architecture configuration, the indication including at least one of receiving an identifier associated with one of a plurality of candidate neural network architecture configurations stored locally on the first device, or receiving one or more data structures representing parameters of the neural network architecture configuration; generating the predicted future CSI estimate includes, in the receiving neural network, generating the predicted future CSI estimate further based on at least one of the following: sensor data input to the receiving neural network from one or more sensors of the first device, or current operating parameters of the RF antenna interface; the method may further include participating in joint training of the neural network architecture configuration of the receiving neural network with the neural network architecture configuration of the transmitting neural network of the second device. The method also includes generating a CSI pilot signal in a transmit neural network of the first device and controlling an RF antenna interface of the first device to transmit a second RF signal representing the CSI pilot signal for reception by the second device. The generating the CSI pilot signal further includes generating the CSI pilot signal in the transmit neural network based on at least one of the following: a carrier frequency of a channel associated with the predicted future CSI estimates, or at least one operating parameter of the RF antenna interface of the first device. The generating the CSI pilot signal further includes generating the CSI pilot signal in the transmit neural network based on at least one of the following:The step of generating the predicted future CSI estimate further includes generating the predicted future CSI estimate based on at least one of the following in the transmit neural network: sensor data from one or more sensors of the first device, a carrier frequency of a channel associated with the predicted future CSI estimate, or at least one operating parameter of an antenna interface of the first device. The at least one MIMO process includes at least one of a beamforming process, a space-time coding process, and a multi-user MIMO process. The at least one capability includes at least one of antenna capability, processing capability, power capability, or sensor capability. The receive neural network includes a deep neural network (DNN).

[0008] According to some embodiments, a computer-implemented method includes receiving capability information from at least one of a first device or a second device, the capability information representing at least one capability of a corresponding one of the first device or the second device; selecting a pair of neural network architectural configurations from a set of candidate neural network architectural configurations based on the capability information, the pair of neural network architectural configurations to be jointly trained to implement a channel state information (CSI) estimation feedback process between the first device and the second device; and transmitting a first indication of the pair of first neural network architectural configurations to the first device for implementation in a transmit neural network of the first device. and transmitting second instructions of a pair of second neural network architecture configurations to the second device for implementation in a receiving neural network of the second device. In various embodiments, the method may further include one or more of the following aspects: The at least one capability includes at least one of antenna capability, processing capability, power capability, or sensor capability. The transmitting neural network and the receiving neural network each include a deep neural network (DNN).

[0009] In some embodiments, an apparatus includes: a network interface; at least one processor coupled to the network interface; and a memory storing executable instructions, the executable instructions configured to operate the at least one processor to perform any of the methods described above and herein.

[0010] The present disclosure may be better understood, and its numerous features and advantages made apparent to those skilled in the art by referencing the accompanying drawings, in which: The use of the same reference symbols in different drawings indicates similar or identical items. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 illustrates an example wireless system that uses a channel state feedback (CSF) neural network approach to characterize a wireless channel, according to some embodiments. [Figure 2] 2 illustrates an exemplary hardware configuration of a user equipment (UE) of the wireless system of FIG. 1 according to some embodiments. [Figure 3] 2 illustrates an exemplary hardware configuration of a base station (BS) of the wireless system of FIG. 1 according to some embodiments. [Figure 4] 2 illustrates an exemplary hardware configuration of management infrastructure components of the wireless system of FIG. 1 according to some embodiments. [Figure 5] FIG. 1 illustrates a machine learning (ML) module using a neural network for use in a CSF neural network method, according to some embodiments. [Figure 6] FIG. 1 illustrates a pair of jointly trained neural networks for processing and transmitting CSI estimates between a UE and a BS, according to some embodiments. [Figure 7] 1 is a flow diagram illustrating an example method for joint training of a set of neural networks to facilitate CSF in a wireless system, according to some embodiments. [Figure 8] 1 is a flow diagram illustrating an example method for feedback of CSI estimates using a set of selected and jointly trained neural networks, according to some embodiments. [Figure 9] 9 is a ladder signaling diagram illustrating an example operation of the method of FIG. 8, according to some embodiments. [Figure 10] FIG. 10 illustrates a selected set of jointly trained neural networks for CSI pilot signaling, CSI estimates, and CSI estimate feedback transmission, according to some embodiments. [Figure 11] FIG. 1 is a flow diagram illustrating an example method for determining a CSI estimate and feeding back the CSI estimate using a neural network, according to some embodiments. [Figure 12] 12 is a ladder signaling diagram illustrating an example operation of the method of FIG. 11 according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0012] Detailed Description Channel state feedback (CSF) is used for signal processing using techniques such as beamforming or space-time coding. In at least one embodiment, to efficiently estimate CSI at the receiving device and provide CSI estimates as CSFs to the transmitting device, the transmitting device and receiving device implement one or more of the pilot transmission process, the CSI estimation process, or the CSI feedback process using jointly trained neural networks. This effectively results in a set of neural networks that are trained to provide processing equivalent to a sequence of conventional CSF stages without having to be specifically designed and tested for each CSF stage. To illustrate, in some embodiments, the CSI pilot transmission process and the CSI estimate process are performed using an algorithmic approach, but the process of returning CSI estimates to the transmitter device relies on the use of a set of jointly trained neural networks, effectively including a transmit (TX) neural network at the receiver device that operates to process CSI estimates of radio frequency (RF) transmissions from the receiver device to the transmitter device in a manner that takes into account current wireless channel conditions and efficiently quantizes or compresses the CSI estimates, and a receive (RX) neural network at the transmitter device that operates to receive and process the over-the-air received output from the TX neural network to recover the CSI estimates, or a representation thereof, for use in one or more MIMO management processes at the transmitter device.

[0013] In another embodiment, neural networks are used for each of the pilot transmission, CSI estimation, and feedback stages. In this approach, the transmitter uses a TX neural network that operates to generate CSI pilot outputs that are transmitted wirelessly to the receiver, and then uses an RX neural network to receive the CSI pilots as inputs and generate corresponding CSI outputs that represent CSI estimates. The receiver further uses the TX neural network to receive the CSI estimates and generate CSF outputs that represent quantized or compressed versions of the CSI estimates. The receiver transmits the CSF outputs to the transmitter, where the RX neural network receives and processes the CSF outputs to generate corresponding CSI estimates that the transmitter can use to manage one or more MIMO processes at the transmitter.

[0014] In either approach, the wireless system can employ joint training of multiple candidate neural network architecture configurations of various neural networks used between the transmitting and receiving devices based on any of a variety of parameters, such as the particular carrier frequency or channel being used, the signal format or protocol, the propagation environment (e.g., characterized by sensor data from various sensors), computing resources, sensor resources, power resources, antenna resources, and other capabilities. Thus, the particular neural network configuration used in each of the transmitting and receiving devices can be selected based on a correlation between the particular configurations of these devices and the parameters used to train the corresponding neural network architecture configuration.

[0015] These and other techniques are described below with reference to a "transmitting device" and a "receiving device." As used herein, a "transmitting device" refers to a device that functions as a primary transmitter for a corresponding channel link, and a "receiving device" refers to a device that functions as a primary receiver for a corresponding channel link. However, this does not mean that a transmitting device cannot also receive RF signals over a channel, or that a receiving device cannot also transmit RF signals over a channel. For example, in a CSF context, a transmitting device is a device that transmits some form of CSI pilot signaling, and a receiving device is a device that receives CSI pilot signaling and determines some form of CSI or CSI estimates from the CSI signaling over the channel. However, this same receiving device typically transmits a representation of the determined CSI back to the transmitting device over the same or another channel, in which case the receiving device acts as a transmitter for transmitting CSI feedback, and the transmitting device is a device that receives the CSI feedback. Furthermore, it should be understood that a device may operate as a transmitter for one channel while operating as a receiver for another channel. For example, for channel characterization of a first channel (e.g., a downlink channel) between a first device and a second device, the first device may operate as a transmitter and the second device may operate as a receiver; simultaneously, or at another time, the first device may operate as a receiver and the second device may operate as a transmitter for channel characterization of a second channel (e.g., an uplink channel) between the first device and the second device, or a second channel (e.g., a sidelink channel) between the second device and a third device.

[0016] 1 illustrates a wireless communication system 100 employing neural network-facilitated channel condition feedback in accordance with some embodiments. As illustrated, the wireless communication system 100 is a cellular network including a core network 102 connected to one or more wide area networks (WANs) 104 or other packet data networks (PDNs), such as the Internet. The wireless communication system 100 further includes at least one base station (BS) 108, each of which supports wireless communication with one or more UEs, such as a UE 110, via RF signaling using one or more applicable radio access technologies (RATs) specified in one or more communication protocols or standards. Thus, the BS 108 serves as a wireless interface between the UE 110, various networks and services provided by the core network 102, and other networks, such as packet-switched (PS) data services, circuit-switched (CS) services, and the like. Conventionally, communication of data or signaling from the BS 108 to the UE 110 is referred to as the “downlink” or “DL,” while communication of data or signaling from the UE 110 to the BS 108 is referred to as the “uplink” or “UL.”

[0017] The BS 108 may use any of a variety of RATs, such as operating as a NodeB (or base transceiver station (BTS)) for the Universal Mobile Telecommunications System (UMTS) RAT (also known as "3G"), as an enhanced NodeB (eNodeB) for the Third Generation Partnership Project (3GPP) Long Term Evolution (LTE) RAT, or as a 5G NodeB ("gNB") for the 3GPP Fifth Generation (5G) New Radio (NR) RAT. The UE 110, in turn, may implement any of a variety of electronic devices operable to communicate with the BS 108 via the appropriate RAT, including, for example, a mobile phone, a cellular-enabled tablet or laptop computer, a desktop computer, a cellular-enabled video game system, a server, a cellular-enabled appliance, a cellular-enabled automotive communication system, a cellular-enabled smart watch or other wearable device, etc.

[0018] Communication of information over the air interface formed between the BS 108 and the UE 110 takes the form of RF signals representing both control plane signaling and user data plane signaling. However, the propagation environment of the channel containing the RF signaling frequently changes due to relatively high frequencies, relatively narrow timing margins, relative motion between the transmitting and receiving devices, the presence or movement of buildings, objects, and other interfering objects, and one or more nearby transmission interference sources. Accordingly, in at least one embodiment, the BS 108 and the UE 110 implement transmitter (TX) and receiver (RX) processing paths that incorporate one or more neural networks (NNs) trained or configured to facilitate one or both of an estimate of CSI or feedback of CSI estimate information from the receiving device to the transmitting device. This verification and associated processes, as described herein, are used to verify downlink channels for RF signaling transmitted by the BS 108 for reception by the UE 110. A downlink channel 112, an uplink channel 114 for RF signaling transmitted by UE 110 for reception by BS 108, or each channel 112, 114. Thus, for downlink channel 112, BS 108 acts as a transmitter and UE 110 acts as a receiver for purposes of CSF, while for uplink channel 114, UE 110 acts as a transmitter and BS 108 acts as a receiver for purposes of CSF.

[0019] With respect to the CSF path 116 of the downlink channel 112, the UE 110 employs a TX processing path 118 having a CSI estimate component 120 and a UE CSF TX DNN 122 (or other neural network) having an output coupled to an RF front-end 124 of the UE 110. The BS 108 employs an RX processing path 126 having a BS CSF RX DNN 128 (or other neural network) having an input coupled to an RF front-end 130 of the BS 108, and a MIMO management component 132 having an input coupled to the output of the BS CSF RX DNN 128.

[0020] During operation, the BS 108 transmits RF signals (not shown) representing CSI pilot signals (also commonly referred to as “training signals”) via the RF front end 130, which are received by the RF front end 124 of the UE 110 and processed by the CSI estimation component 120 to generate one or more CSI estimates 134 for each frequency or subcarrier represented by the received CSI pilot signals. The CSI estimation component 120 may use any of a variety of well-known or proprietary techniques to generate at least one CSI estimate 134 from a corresponding set of one or more CSI pilot signals transmitted by the BS 108. For example, if the channel and noise distribution are unknown, the CSI estimation component 120 may determine the CSI estimate 134 using, for example, any of a variety of least-squares estimators. On the other hand, if the channel and noise distribution are known, the CSI estimation component 120 may determine the CSI estimate 134 using, for example, any of a variety of Bayesian estimation techniques. The CSI estimate 134 may take any of a variety of forms or representations known in the art, but for ease of reference, the CSI estimate 134 is implemented as a set of one or more matrices, each matrix representing a corresponding transmission format for a corresponding carrier frequency. However, it will be understood that the CSI estimate 134 is not limited to this particular implementation and may represent any of a variety of suitable CSI estimate formats.

[0021] The CSI estimate 134 is calculated based on the UE CSF along with any of a variety of other inputs. The CSF estimates 134 are provided as inputs to the UE 110, such as sensor data inputs indicative of the current propagation environment as observed by one or more sensors of the UE 110 (and as described below). In at least one embodiment, the UE CSF TX DNN 122 is trained together with the BS CSF RX DNN 128 of the BS 108, and thus generates CSF outputs from the CSI estimates 134 (and other inputs) that are suitable for RF transmission to the BS 108 and processing by the BS CSF RX DNN 128. As part of this joint training or other configuration, the UE CSF TX DNN 122, in at least one embodiment, is trained or configured to effectively quantize or compress the data or information represented by the CSI estimates 134 and otherwise process the resulting compressed information given the input sensor data or other inputs to generate the CSF output 136 that is provided to the RF front end 124 for wireless transmission to the BS 108.

[0022] At the BS 108, the RF front end 130 extracts the CSF output 136 from the received RF signaling and uses the CSF output 136 as an input to the BS CSF RX DNN 128. Optional other inputs, such as sensor data from sensors at the BS 108, are also provided to the BS The BS 108 may simultaneously provide the CSI estimates 134 as inputs to the CSF RX DNN 128. From these inputs and based on joint training or other configurations, the BS CSF RX DNN 128 operates to provide a recovered representation of the CSI estimates 134, referred to herein as recovered CSI estimates 138. The recovered CSI estimates 138 are then provided to the MIMO management component 132, which uses the recovered CSI estimates 138 to control one or more MIMO processes of the BS 108 for downlink channel 112 with the UE 110, such as by controlling one or more of a beamforming process, a space-time coding process, or a multi-user MIMO process utilized by the RF front end 130 for RF transmission to at least the UE 110.

[0023] While FIG. 1 illustrates a CSF path 116 for a downlink channel 112, it will be appreciated that a configuration similar to that illustrated can also be utilized to provide a CSF path for an uplink channel 114, with the BS 108 using a similar CSI estimation component and CSF TX DNN and the UE 110 using a similar CSF RX DNN and MIMO management component. For ease of explanation, however, the neural network-based CSF techniques of this disclosure will be described in the context of an example downlink channel from a BS to a UE. However, it will be understood that these same techniques are equally applicable to an uplink channel from a UE to a BS, or an uplink channel between any RF-enabled device acting as a transmitter and another RF-enabled device acting as a receiver for CSF purposes. Furthermore, FIG. 1 illustrates an example implementation in which the generation and transmission of CSI pilot signals by a transmitter and algorithmic estimation of CSI based on the CSI pilot signals received at a receiver can be performed using conventional approaches. In other embodiments, one or more of these processes may be implemented partially or fully using co-trained neural networks, as described below with reference to Figures 10-12.

[0024] As noted above and described in more detail herein, both the transmitting and receiving devices (e.g., the BS 108 and the UE 110, respectively, in the case of the CSF path 116) use one or more DNNs or other neural networks that are jointly trained and selected based on context-specific parameters to facilitate the overall CSF process. To manage the joint training, selection, and maintenance of these neural networks, in at least one embodiment, the system 100 further includes a management infrastructure component 140 (or, for brevity, a “management component 140”). This management component 140 may include a server or other component within the network infrastructure 106 of the wireless communication system 100, such as, for example, within the core network 102 or the WAN 104. Furthermore, while shown as separate components in the illustrated example, in at least some embodiments, the BS 108 implements the management component 140. The monitoring functions provided by management component 140 may include, for example, some or all of the following: overseeing the joint training of neural networks; managing the selection of a particular neural network architecture configuration for one or more of the sending or receiving devices based on specific capabilities or other component-specific parameters; receiving and processing capability updates for purposes of selecting a neural network configuration; receiving and processing feedback for purposes of training or selecting a neural network;

[0025] As will be described in more detail below, in some embodiments, the management component 140 maintains a set 142 of candidate neural network architecture configurations 144, and determines the current capabilities of components that implement the corresponding neural networks, A channel may be selected for use by a particular component in the corresponding CSF path based at least in part on the current capabilities of other components in the BS 108, the UE 110, or a combination thereof. These capabilities may include, for example, sensor capabilities, processing resource capabilities, battery / power capabilities, RF antenna capabilities, capabilities of one or more accessories of the component, the type or nature of data transmitted over the corresponding channel, etc. Information representing these capabilities of the BS 108 and the UE 110 is obtained by management component 140 and stored as BS capability information 146 and UE capability information 148, respectively. Management component 140 may further consider parameters of the corresponding channel or propagation channel or other aspects of the environment, such as the channel's carrier frequency, the known presence of objects or other sources of interference, etc. Information representing aspects of the channel or propagation environment is obtained by management component 140 and stored as channel / propagation information 150.

[0026] To support this approach, in some embodiments, management component 140 can manage the joint training of different combinations of candidate neural network architecture configurations 144 for different capability / context combinations. Management component 140 can then obtain capability information 146 from BS 108, capability information 148 from UE 110, or both, from which management component 140 selects a neural network architecture configuration from set 142 of candidate neural network architecture configurations 144 for each component based at least in part on the corresponding indicated capabilities and the RF signal propagation environment reflected in channel / propagation information 150. In some embodiments, the candidate neural network architecture configurations are jointly trained as paired subsets, such that each candidate neural network architecture configuration for a particular capability set of BS 108 is jointly trained with a single corresponding candidate neural network architecture configuration for a particular capability set of UE 110. In other embodiments, candidate neural network architecture configurations are trained such that each candidate configuration for BS 108 has a one-to-many correspondence with multiple candidate configurations for UE 110, and vice versa.

[0027] Thus, system 100 utilizes a CSF approach that relies on a set of neural networks that are managed, jointly trained, and selectively used between the transmitter and receiver devices for CSI feedback, rather than independently designed process blocks that may not be specifically designed for compatibility. This not only increases flexibility but also, in some circumstances, allows for faster processing and more efficient RF transmission at each device, thereby reducing latency in the estimation, communication, and implementation of CSI estimates. This, in turn, facilitates finer and more timely control of the MIMO process, resulting in more efficient and effective signaling between the transmitter and receiver devices.

[0028] 2 illustrates an example hardware configuration of a UE 110 (as a representative receiving device) according to some embodiments. Note that the depicted hardware configuration represents the processing and communication components most directly related to the neural network-based processes described herein, and omits certain well-understood components that are frequently implemented in such electronic devices, such as a display, non-sensor peripherals, and an external power source.

[0029] In the illustrated configuration, the UE 110 includes an RF front end 124 having one or more antennas 202 and an RF antenna interface 204 having one or more modems supporting one or more RATs. The RF front end 124 effectively acts as a physical (PHY) transceiver interface that performs and processes signaling between one or more processors 206 of the UE 110 and the antennas 202 to facilitate various types of wireless communication. The antennas 202 may be similar or different from one another. The UE 110 may be arranged in one or more arrays of multiple antennas configured to transmit and receive signals and may be tuned to one or more frequency bands associated with the corresponding RAT. The one or more processors 206 may include, for example, one or more central processing units (CPUs), graphics processing units (GPUs), tensor processing units (TPUs), or other application-specific integrated circuits (ASICs). To illustrate, the processor 206 may include an application processor (AP) utilized by the UE 110 to execute operating systems and various user-level software applications, as well as one or more processors utilized by a baseband processor of the modem or RF front-end 124. The UE 110 further includes one or more computer-readable media 208, including any of a variety of media used by electronic devices to store data and / or executable instructions, such as random access memory (RAM), read-only memory (ROM), cache, flash memory, solid-state drive (SSD), or other mass storage device. For ease of explanation and brevity, the computer-readable medium 208 will be referred to herein as "memory 208" given that system memory or other memory is frequently used to store data and instructions executed by the processor 206, although it will be understood that references to "memory 208" shall apply to other types of storage media as well unless otherwise specified.

[0030] In at least one embodiment, the UE 110 further includes a plurality of sensors, referred to herein as sensor set 210, at least some of which are utilized in the neural network-based manner described herein. Generally, the sensors in sensor set 210 include sensors that sense some aspect of the UE 110's environment or a user's use of the UE 110, potentially sensing parameters that have at least some impact on or reflect the UE 110's RF propagation path relative to the BS 108 or RF transmission and reception performance by the UE 110. The sensors in sensor set 210 may include one or more sensors for object detection, such as a radar sensor, a lidar sensor, an imaging sensor, a structured light-based depth sensor, etc. The sensor set 210 may also include one or more sensors for determining the position or attitude of the UE 110, such as a satellite positioning sensor, such as a GPS sensor, a global navigation satellite system (GNSS) sensor, an internal measurement unit (IMU) sensor, a visual odometry sensor, a gyroscope, a tilt sensor or other inclinometer, an ultra-wideband (UWB)-based sensor, etc. Other examples of types of sensors in sensor set 210 may include imaging sensors, e.g., cameras for image capture by a user, cameras for face detection, cameras for stereoscopic vision or visual odometry, light sensors for detecting objects near features of the device, etc. UE 110 may further include one or more batteries 212 or other portable power source, as well as one or more user interface (UI) components 214, which may include, for example, a touchscreen, user-operable input / output devices (e.g., “buttons” or keyboard) or other touch / contact sensors, a microphone or other audio sensor for capturing audio content, an image sensor for capturing video content, a thermal sensor (e.g., to detect proximity to a user), etc.

[0031] The one or more memories 208 of the UE 110 are used to store one or more sets of executable software instructions and associated data that operate the one or more processors 206 and other components of the UE 110 to perform the various functions described herein and attributed to the UE 110. The sets of executable software instructions include, for example, an operating system (OS) and various drivers (not shown), as well as various software applications. The sets of executable software instructions may also be used to manage the neural network management module 222, the performance management module 224, or the CSI estimation module 226 (one implementation of the CSI estimation component 120 of FIG. 1). The UE 110 may further include one or more of the following: a neural network management module 222 implements one or more neural networks for the UE 110, as described in detail below. The capability management module 224 determines various capabilities of the UE 110 that may be related to neural network configuration or selection, and reports such capabilities to the management component 140. Similarly, the UE 110 monitors the UE 110 for changes in such capabilities, including changes in RF and processing capabilities, changes in accessory availability or capabilities, and the like, and manages reporting of such capabilities and changes to the management component 140. Also as described above, the CSI estimation module 226 operates to generate CSI estimates based on received representations of CSI pilot signals transmitted by another device, such as the BS 108. In at least one embodiment, the CSI estimation module 226 implements one or more algorithmic techniques for calculating CSI estimates, such as any of various least squares (LS), least mean squares (LMS), or Bayesian CSI estimation techniques known in the art.

[0032] To facilitate the operations of the UE 110 described herein, the one or more memories 208 of the UE 110 may further store data associated with these operations. This data may include, for example, device data 228 and one or more neural network architecture configurations 230. The device data 228 may represent, for example, user data, multimedia data, beamforming codebooks, software application configuration information, etc. The device data 228 may further include capability information of the UE 110, such as sensor capability information for one or more sensors of the sensor set 210. This information may include the presence or absence of a particular sensor or sensor type, and, for sensors present, one or more representations of corresponding capabilities, such as the range and resolution of a lidar or radar sensor, the image resolution and color depth of an imaging camera, etc. The capability information may further include information regarding, for example, the capability or status of the battery 212, the capability or status of the UI 214 (e.g., the screen resolution, color gamut, or frame rate of a display), etc.

[0033] One or more neural network architecture configurations 230 represent UE implementations selected from a set 142 of candidate neural network architecture configurations 144 maintained by management component 140. Each neural network architecture configuration 230 includes one or more data structures containing data and other information representing a corresponding architecture and / or parameter configuration used by neural network management module 222 to form a corresponding neural network for UE 110. Information included in neural network architecture configuration 230 may include, for example, parameters specifying a fully connected layer neural network architecture, a convolutional layer neural network architecture, recurrent neural network layers, several connected hidden neural network layers, an input layer architecture, an output layer architecture, several nodes utilized by the neural network, coefficients (e.g., weights and biases) utilized in the neural network, kernel parameters, several filters utilized by the neural network, stride / pooling configurations utilized by the neural network, activation functions for each neural network layer, interconnections between neural network layers, neural network layers to skip, etc. Thus, neural network architecture configuration 230 includes any combination of NN formation components (e.g., architecture and / or parameter configurations) that can be used to create a NN formation configuration (e.g., a combination of one or more NN formation components) to define and / or form a DNN.

[0034] 3 illustrates an example hardware configuration for BS 108 (as a representative transmitting device) according to some embodiments. The depicted hardware configuration includes the processing and communication components most directly related to the neural network-based processes described herein. It should be noted that the illustrated diagram depicts an implementation of the BS 108 as a single network node (e.g., a 5G NR Node B, or "gNB"), omitting certain components that are well understood to be frequently implemented in such electronic devices, such as a display, non-sensor peripherals, an external power source, etc. Additionally, it should be noted that while the illustrated diagram depicts an implementation of the BS 108 as a single network node (e.g., a 5G NR Node B, or "gNB"), the functionality, and therefore alternatively the hardware components of the BS 108, may be distributed across multiple network nodes or devices, and may be distributed in a manner that performs the functions described herein.

[0035] In the illustrated configuration, the BS 108 includes an RF front end 130 having one or more antennas 302 and an RF antenna interface 304 having one or more modems supporting one or more RATs, which act as a PHY transceiver interface that performs and processes signaling between one or more processors 306 of the BS 108 and the antennas 302 to facilitate various types of wireless communications. The antennas 302 may be arranged in one or more arrays of multiple antennas configured similarly or differently from one another and may be tuned to one or more frequency bands associated with corresponding RATs. The one or more processors 306 may include, for example, one or more CPUs, GPUs, TPUs, or other ASICs. The BS 108 further includes one or more computer-readable media 308, which may include any of a variety of media used by electronic devices to store data and / or executable instructions, such as RAM, ROM, cache, flash memory, SSD, or other mass storage device. Similar to the memory 208 of the UE 110, for ease of explanation and brevity, the computer-readable medium 308 will be referred to herein as “memory 308” in light of the frequent use of system memory or other memory for storing data and instructions for execution by the processor 306. However, it will be understood that reference to “memory 308” should apply to other types of storage media as well, unless otherwise specified.

[0036] In at least one embodiment, the BS 108 further includes a plurality of sensors, referred to herein as sensor set 310, at least some of which are utilized in the neural network-based approach described herein. Generally, the sensors in sensor set 310 include sensors that sense some aspect of the BS 108's environment, potentially sensing parameters that have at least some impact on or reflect the BS 108's RF propagation path to the corresponding UE 110 or the RF transmission and reception performance by the BS 108. The sensors in sensor set 310 may include one or more sensors for object detection, such as a radar sensor, a lidar sensor, an imaging sensor, a structured light-based depth sensor, etc. If the BS 108 is a mobile BS, the sensor set 310 may also include one or more sensors for determining the BS 108's position or attitude. Other examples of types of sensors in sensor set 310 may include imaging sensors, optical sensors for detecting objects in proximity to features of the BS 108, etc.

[0037] The one or more memories 308 of the BS 108 are used to store one or more sets of executable software instructions and associated data that operate the one or more processors 306 and other components of the BS 108 to perform the various functions described herein and attributed to the BS 108. The sets of executable software instructions include, for example, an operating system (OS) and various drivers (not shown), as well as various software applications. The sets of executable software instructions further include one or more of a neural network management module 314, a capability management module 316, a CSF management module 318, or a MIMO management module 320.

[0038] The neural network management module 314 implements one or more neural networks for the BS 108, as described in detail below. The capability management module 318 determines various capabilities of the BS 108, which may be related to neural network configuration or selection, reports such capabilities to the management component 140, and similarly monitors the BS 108 for changes in such capabilities, including changes in RF and processing capabilities, and manages such capability reporting and capability changes to the management component 140. The CSF management module 318 operates to manage CSF processes between the BS 108 and one or more corresponding UEs 110, including managing the generation and transmission of CSI pilot signals to the corresponding UEs 110, obtaining and processing resulting CSI estimate information reported by the UEs 110 from analysis of the transmitted CSI pilot signals, and communicating a representation of the resulting CSI estimates to the MIMO management module 320. The MIMO management module 320 then operates to control one or more MIMO processes of the RF front end 124 based on the provided CSI estimates. These MIMO processes may include beamforming processes, space-time coding processes, MU-MIMO processes, etc.

[0039] To facilitate the operation of the BS 108 as described herein, the one or more memories 308 of the BS 108 may further store data associated with these operations. This data may include, for example, BS data 328 and one or more neural network architecture configurations 330. The BS data 328 may represent, for example, a beamforming codebook, software application configuration information, etc. The BS data 328 may also include capability information of the BS 108, such as sensor capability information for one or more sensors of the sensor set 310, including the presence or absence of a particular sensor or sensor type and, for present sensors, one or more representations of corresponding capabilities, such as the range and resolution of a lidar or radar sensor, the image resolution and color depth of an imaging camera, etc. The one or more neural network architecture configurations 330 represent BS implementation examples selected from the set 142 of candidate neural network architecture configurations 144 maintained by the management component 140. Thus, similar to the neural network architecture configuration 230 of FIG. 2 , each neural network architecture configuration 330 includes one or more data structures containing data and other information representing the corresponding architecture and / or parameter configuration used by the neural network management module 314 to form the corresponding neural network of the BS 108.

[0040] 4 illustrates an example hardware configuration for management component 140 according to some embodiments. Note that the depicted hardware configuration represents the processing and communication components most directly related to the neural network-based processes described herein, and omits certain components that are well understood to be frequently implemented in such electronic devices. Additionally, while the hardware configuration is shown as being located in a single component, the functionality of management component 140, and therefore the hardware components, may instead be distributed across multiple infrastructure components or nodes, and may be distributed in a manner that performs the functions described herein.

[0041] As noted above, the management component 140 may be implemented in any of a variety of components or combinations of components within the network infrastructure 106. For ease of explanation, the management component 140 is described herein with reference to an example implementation as a server or other component of one of the core network 102, although in other embodiments, the management component 140 may be implemented as part of the BS 108, for example.

[0042] As shown, management component 140 includes one or more network interfaces 402 (e.g., Ethernet interfaces) for coupling to one or more networks of system 100, one or more processors 404 coupled to the one or more network interfaces 402, and one or more non-transitory computer-readable storage media 406 (referred to herein as “memory 406” for brevity) coupled to the one or more processors 404. The one or more memories 406 store one or more sets of executable software instructions and associated data that operate the one or more processors 404 and other components of management component 140 to perform the various functions described herein and attributed to management component 140. The sets of executable software instructions include, for example, an OS and various drivers (not shown). The software stored in the one or more memories 406 may further include one or more of a training module 408 or a neural network selection module 410. The training module 408 operates to manage the joint training of candidate neural network architecture configurations 144 for a set of candidate neural networks 142 that can be used at the transmitting and receiving devices in the CSF path using one or more sets of training data 416. Training can include training the neural networks offline (i.e., while not actively involved in processing communications) and / or online (i.e., while actively involved in processing communications). Furthermore, training can be performed individually or separately, such that each neural network is trained individually based on its own training data set without the results being communicated to or affecting the DNN training on the other side of the transmission path, or the training can be joint training, such that neural networks in a data stream transmission path are jointly trained on the same or complementary data sets.

[0043] 1, the neural network selection module 410 operates to obtain, filter, and otherwise process selection-related information 420 from one or both of the transmitting and receiving devices in the CSF path, such as the BS 108 and the UE 110, respectively, and uses this selection-related information 420 to select a pair of jointly trained neural network architecture configurations 144 from the candidate set 142 for implementation at the transmitting and receiving devices in the CSF path. As described above, this selection-related information 420 may include, for example, current capability information from one or both of the UE 110 and the BS 108, current propagation path information, channel-specific parameters, etc. After the selection is made, the neural network selection module 410 begins transmitting an indication of the selected neural network architecture configuration 144 to each network component, for example, via transmitting an index number associated with the selected configuration, transmitting one or more data structures representing the neural network architecture configuration itself, or a combination thereof.

[0044] 5 illustrates an example machine learning (ML) module 500 for implementing a neural network according to some embodiments. As described herein, one or both of the transmitting and receiving devices in the CSF path (e.g., the BS 108 and the UE 110, respectively, in the CSF path 116 of FIG. 1) implement one or more DNNs or other neural networks to process incoming or outgoing wireless communications related to CSI feedback. Accordingly, the ML module 500 illustrates an example module for implementing one or more of these neural networks.

[0045] In the illustrated example, the ML module 500 is a set of connected nodes organized into three or more layers. The ML module 500 implements at least one deep neural network (DNN) 502 having a group of neural networks (e.g., neurons and / or perceptrons). The nodes between layers can be configured in various ways, such as a partially connected configuration in which a first subset of nodes in a first layer are connected to a second subset of nodes in a second layer, or a fully connected configuration in which each node in the first layer is connected to each node in the second layer. Neurons process input data and generate continuous output values, such as real numbers between 0 and 1. In some cases, the output value indicates how close the input data is to a desired category. Perceptrons perform linear classification, such as binary classification, on the input data. Nodes, whether neurons or perceptrons, can use a variety of algorithms to generate output information based on adaptive learning. Using the DNN 502, the ML module 500 performs a variety of different types of analysis, including single linear regression, multiple linear regression, logistic regression, stepwise regression, binary classification, multi-class classification, multivariate adaptive regression splines, locally estimated scatterplot smoothing, and the like.

[0046] In some implementations, the ML module 500 adaptively learns based on supervised learning. In supervised learning, the ML module 500 receives various types of input data as training data. The ML module 500 processes the training data to learn how to map the input to a desired output. As one example, the ML module 500 receives digital samples of a signal as input data and learns how to map the signal samples to binary data that reflects information embedded within the signal. As another example, the ML module 500 receives binary data as input data and learns how to map the binary data to digital samples of the signal where the binary data is embedded within the signal. As yet another example, when used in TX mode, as described in more detail below, the ML module 500 receives an outgoing information block and learns both how to generate an output that actually represents the data to be encoded (e.g., compressed) and the channel to be encoded represented within the information block, forming an output suitable for wireless transmission over an RF antenna interface. Conversely, when implemented in RX mode, the ML module 500 can be trained to receive inputs that effectively represent data-encoded and channel-encoded representations of information blocks, process inputs to generate outputs that are effectively data-decoded and channel-decoded representations of the inputs, and thus represent data-recovered representations of the information. As described further below, training in either or both of the TX mode or the RX mode can further include training using sensor data as inputs, capability information as inputs, accessory information as inputs, RF antenna configurations, or other operational parameter information as inputs, etc.

[0047] During the training procedure, the ML module 500 uses labeled or known data as input to the DNN 502. The DNN 502 analyzes the input using nodes and generates corresponding outputs. The ML module 500 compares the corresponding outputs with the truth data and adapts the algorithms implemented by the nodes to improve the accuracy of the output data. The DNN 502 then applies the adapted algorithm to the unlabeled input data to generate corresponding output data. The ML module 500 uses one or both of statistical analysis and adaptive learning to map inputs to outputs. For example, the ML module 500 uses characteristics learned from the training data to correlate unknown inputs with outputs that are statistically likely to be within a threshold range or value. This allows the ML module 500 to receive complex inputs and identify corresponding outputs. As mentioned above, some implementations train the ML module 500 on characteristics of communications transmitted over wireless communication systems (e.g., time / frequency interleaving, time / frequency deinterleaving, convolutional coding, convolutional decoding, power levels, channel equalization, inter-symbol interference, quadrature amplitude modulation / demodulation, frequency division multiplexing / demultiplexing, transmission channel characteristics) as well as characteristics of data encoding / decoding schemes used in such systems. This allows the trained ML module 500 to receive as input samples of a signal and to generate a signal. It is possible to recover information such as binary data embedded in a signal from the signal.

[0048] In the illustrated example, the DNN 502 includes an input layer 504, an output layer 506, and one or more hidden layers 508 located between the input layer 504 and the output layer 506. Each layer may have any number of nodes, and the number of nodes between layers may be the same or different. That is, the input layer 504 may have the same and / or a different number of nodes as the output layer 506, the output layer 506 may have the same and / or a different number of nodes as the one or more hidden layers 508, etc.

[0049] Node 510 corresponds to one of several nodes included in input layer 504, and the node performs separate and independent calculations. As will be further described, the node receives input data and processes the input data using one or more algorithms to generate output data. Typically, the algorithm includes weights and / or coefficients that change based on adaptive learning. Thus, the weights and / or coefficients reflect information learned by the neural network. Each node may optionally determine whether to pass the input data to one or more subsequent nodes for processing. Specifically, after processing the input data, node 510 may determine whether to pass the input data to one or both of node 512 and node 514 in hidden layer 508 for processing. Alternatively or additionally, node 510 passes the input data to a node for processing based on the layer connection architecture. This process may be repeated across multiple layers until DNN 502 generates output using a node in output layer 506 (e.g., node 516).

[0050] Neural networks can also use various architectures that determine which nodes within the neural network are connected, how data is advanced and / or retained within the neural network, what weights and coefficients are used to process input data, how the data is processed, and so on. These various elements collectively describe a neural network architectural configuration, such as the neural network architectural configuration briefly described above. To explain, recurrent neural networks, such as long short-term memory (LSTM) neural networks, form cycles between node connections to retain information from previous portions of an input data sequence. Recurrent neural networks use the retained information for subsequent portions of the input data sequence. As another example, feedforward neural networks pass information to forward connections without forming cycles that retain information. While described in the context of node connections, it should be understood that neural network architectural configurations can include various parameter configurations that affect how the DNN 502 or other neural network processes input data.

[0051] The neural network architecture configuration of a neural network can be characterized by various architecture and / or parameter configurations. For illustrative purposes, consider an example in which the DNN 502 implements a convolutional neural network (CNN). Generally, a convolutional neural network corresponds to a type of DNN in which layers process data using convolution operations to filter input data. Thus, the CNN architecture configuration can be characterized by, for example, pooling parameters, kernel parameters, weights, and / or layer parameters.

[0052] Pooling parameters correspond to parameters that specify a pooling layer in a convolutional neural network that reduces the dimensionality of input data. To illustrate, a pooling layer can combine the outputs of nodes in a first layer with the inputs of nodes in a second layer. Alternatively or additionally, pooling parameters specify where and how the neural network pools data in the data processing layers. A pooling parameter indicating "max pooling" is: For example, a neural network may be configured to select and pool the maximum values ​​from a group of data generated by nodes in a first layer and use the maximum values ​​as input to a single node in a second layer. A pooling parameter indicating "average pooling" configures a neural network to generate an average value from a group of data generated by nodes in the first layer and use the average value as input to a single node in the second layer.

[0053] Kernel parameters indicate the filter size (e.g., width and height) to use when processing input data. Alternatively or additionally, kernel parameters specify the type of kernel method used to filter and process the input data. For example, a support vector machine corresponds to a kernel method that uses regression analysis to identify and / or classify data. Other types of kernel methods include Gaussian processes, canonical correlation analysis, spectral clustering methods, etc. Thus, kernel parameters can indicate the filter size and / or type of kernel method to apply to a neural network. Weight parameters specify the weights and biases used by an algorithm in a node to classify input data. In some implementations, the weights and biases are learned parameter configurations, such as parameter configurations generated from training data. Layer parameters specify layer connections and / or layer types, such as a fully connected layer type indicating that all nodes in a first layer (e.g., output layer 506) are connected to all nodes in a second layer (e.g., hidden layer 508), a partially connected layer type indicating which nodes in the first layer are disconnected from the second layer, and an activation layer type indicating which filters and / or layers are activated within the neural network. Alternatively or additionally, the layer parameters specify a type of node layer, such as a normalization layer type, a convolutional layer type, a pooling layer type, etc.

[0054] Although described in the context of pooling parameters, kernel parameters, weight parameters, and layer parameters, it will be understood that other parameter configurations can be used to form a DNN consistent with the guidelines provided herein. Thus, a neural network architecture configuration can include any suitable type of configuration parameters applicable to a DNN that affect how the DNN processes input data and generates output data.

[0055] In some embodiments, the configuration of the ML module 500 is further based on the current operating environment. To illustrate this, consider an ML module trained to generate binary data from digital samples of a signal. The RF signal propagation environment often changes the characteristics of a signal as it passes through a physical environment. RF signal propagation environments frequently change, affecting how the environment alters the signal. For example, a first RF signal propagation environment alters the signal in a first way, while a second RF signal propagation environment alters the signal in a different way. These differences affect the accuracy of the output results generated by the ML module 500. For example, a DNN 502 configured to process communications transmitted in a first RF signal propagation environment may generate errors or limit performance when processing communications transmitted in a second RF signal propagation environment. Particular sensors in a sensor set of a component implementing the DNN 502 can provide sensor data representative of one or more aspects of the current RF signal propagation environment. Examples of the above may include lidar, radar, or other object detection sensors for determining the presence or absence of interfering objects in the line-of-sight propagation path, UI sensors for determining the presence and / or location of a user's body relative to the component, etc. However, it will be appreciated that the particular sensor capabilities available may depend on the particular components implementing the sensors. For example, the BS 108 may have lidar or radar capabilities and therefore the ability to detect nearby objects, while the UE 110 may lack lidar and radar capabilities. As another example, a smartphone (one embodiment of the UE 110) may be able to sense whether the smartphone is in a user's pocket or bag. A device such as a notebook computer (another embodiment of UE 110) may have a light sensor that can be used to sense the temperature, whereas a notebook computer (another embodiment of UE 110) may lack this capability. Thus, in some embodiments, the particular configuration implemented for ML module 500 may depend at least in part on the particular sensor configuration of the device implementing ML module 500.

[0056] The architectural configuration of the ML module 500 may also be based on the capabilities of the node implementing the ML module 500, the capabilities of one or more nodes upstream or downstream of the node implementing the ML module 500, or a combination thereof. For example, the UE 110 may be battery power limited, and thus the ML modules 500 of both the UE 110 and the BS 108 may be trained based on battery power as an input, for example, facilitating the ML modules 500 at both ends to use a CSI estimate coding scheme that is more suitable for lower power consumption. Furthermore, in some embodiments, the architectural configuration of the ML module 500, when implemented in a TX processing module for CSI feedback, may be based on or trained for a prediction of CSI in a future period when the CSI is considered for control of one or more MIMO processes (predicted future CSI estimates), and thus the ML module 500 may be trained to use such a prediction.

[0057] Thus, in some embodiments, a device implementing ML module 500 may be configured to implement different neural network architecture configurations for different combinations of capability parameters, RF environment parameters, etc. For example, the device may have access to a different set of one or more neural network architecture configurations to use when an imaging camera is available on the device and other devices in the CSF path utilize lidar, and one or more neural network architecture configurations to use when an imaging camera is not available on the device and other devices utilize radar.

[0058] In some embodiments, a device implementing ML module 500 locally stores some or all of a set of candidate neural network architectural configurations available for ML module 500. For example, the candidate neural network architectural configurations may be indexed by a look-up table (LUT) or other data structure that receives one or more parameters as input, such as one or more capability parameters, propagation environment parameters, one or more channel parameters, etc., and outputs an identifier associated with the corresponding locally stored candidate neural network architectural configuration that is appropriate for operation given the input parameters. However, in some embodiments, the neural network used by the transmitting device and the neural network used by the receiving device are trained jointly, and therefore, a mechanism may need to be used between the transmitting device and the receiving device to ensure that each device selects a jointly trained neural network architectural configuration for its ML module 500, or at least that it is operationally compatible with the neural network architectural configuration selected by the other device for its complementary ML module 500. This mechanism may include, for example, coordination signaling sent between the two devices directly or via the management component 140, or the management component 140 may act as a criterion for selecting a pair of compatible co-trained architecture configurations from the subsets proposed by each device.

[0059] However, in other embodiments, it may be more efficient or otherwise advantageous to have the management component 140 operate to select a pair of appropriate co-trained neural network architecture configurations for use in the corresponding ML modules 500 of the transmitting and receiving devices. In this approach, the management component 140 may receive information from the transmitting and receiving devices that represents some or all of the parameters that may be used in the selection process. ML module 500 of that device, or alternatively, management component 140 may transmit one or more data structures representing the neural network architecture configuration selected for that device.

[0060] To facilitate the process of selecting a pair of appropriate neural network architecture configurations for the transmitting and receiving devices, in at least one embodiment, the management component 140 trains the ML module 500 in a training CSF path using an appropriate combination of a neural network management module and a training module. Training can be performed offline when no active communication exchange is taking place, or online during an active communication exchange. For example, the management component 140 can mathematically generate training data, access files that store training data, acquire real-world communication data, etc. The management component 140 then extracts and stores the various learned neural network architecture configurations for subsequent use. In some implementations, input characteristics are stored with each neural network architecture configuration, whereby the input characteristics describe various characteristics of the RF signal propagation environment and / or performance configuration corresponding to the respective neural network architecture configuration. In the implementation, the neural network manager selects the neural network architecture configuration by matching the current RF signal propagation environment and the current operating environment to the input characteristics, where the current operating environment includes an indication of the capabilities of one or more nodes along the training CSF path, such as sensor capability, RF capability, streaming accessory capability, processing capability, scheduling latency, etc.

[0061] As previously described, wirelessly communicating network devices, such as the BS 108 and the UE 110, can be configured to process wireless communication exchanges using one or more DNNs at each network device, with each DNN replacing one or more functions traditionally implemented by one or more hard-coded or fixed-design blocks to facilitate the CSI estimation or CSF process. Additionally, each DNN can further incorporate current sensor data from one or more sensors in the networked device's sensor set and / or capability data from some or all of the nodes along the chain 116, effectively modifying or adapting its operation to the current operating environment.

[0062] To this end, Figure 6 illustrates an example operating environment 600 for a DNN implementation in the example CSF path 116 of Figure 1, with the BS 108 acting as a transmitting device and the UE 110 acting as a receiving device. In the illustrated example, the neural network management module 222 of the UE 110 implements a CSF transmitter (TX) processing module 602, while the neural network management module 314 of the BS 108 implements a CSF receiver (RX) processing module 604. In at least one embodiment, each of these processing modules implements one or more DNNs via implementation of a corresponding ML module, as described above with reference to the one or more DNNs 502 of the ML module 500 of Figure 5.

[0063] In the illustrated approach, the operating environment 600 utilizes a conventional approach to the process of generating CSI estimates, but does not utilize a conventional approach to encoding and transmitting the CSI estimates to the BS 108. utilizes a neural network-based approach. Accordingly, the CSF management module 318 generates a sequence of one or more CSI pilot signals 608, each of which is provided to the RF antenna interface 304 of the BS 108 for conversion into a corresponding RF signal 610 that is transmitted to the UE 110 via one or more antennas 302. The RF signals 610 are received and processed at the UE 110 via one or more antennas 202 and the RF antenna interface 204, and the resulting acquired signals 612 are analyzed by the CSI estimation module 226 to generate one or more CSI estimates for the antennas / receivers / subcarriers corresponding to the transmitted CSI pilot signals 608. The CSI estimation module 226 can use any of a variety of well-known or proprietary techniques for CSI estimation. Generally, such CSI estimation techniques leverage the fact that parameters of the CSI pilot signals 608 are known a priori, and thus can compare the actual received form of the acquired signal 612 with its expected received form to determine a CSI estimate.

[0064] In a conventional CSF process, the CSI estimate for a corresponding antenna / receiver / subcarrier combination is added to the corresponding index location of a CSI estimate matrix, which, once complete, is transmitted back from the UE to the BS using a fixed, hard-coded compression algorithm such as vector quantization (VQ) to reduce the total amount of data transmitted for the CSI estimate matrix. However, in complex systems such as massive MIMO or MU-MIMO systems, the size of the resulting CSI estimate matrix increases exponentially as the number of antennas, subcarriers, and users increases, which can make conventional hard-coded approaches to quantizing or compressing the CSI estimate matrix for transmission infeasible or at least excessively complex.

[0065] As such, instead of using a hard-coding or algorithmic CSI estimate quantization process, the CSF TX processing module 602 of the UE 110 and the CSF RX processing module 604 of the BS 108 instead interoperate to support a neural network-based wireless feedback path between the UE 110 and the BS 108 to communicate data representing CSI estimates determined by the UE 110 from CSI pilot signaling transmitted by the BS 108. To this end, the CSF of the UE 110 One or more DNNs in the TX processing module 602 are trained to receive as input an output CSI estimate data block 614 representing one or more CSI estimates generated by the CSI estimate module 226, as well as other inputs, and generate corresponding CSI outputs 620 from these inputs.

[0066] Optional other inputs provided to the CSF TX processing module 602 may include, for example, sensor data 616 from the sensor set 210 and network status information 618 representing current operating parameters of the transmit side of the RF antenna interface 204, and thus serving as a representation of the current RF propagation environment for transmitting signaling. Furthermore, it will be understood that channel conditions may change from time to time, and thus a latency (referred to herein as “scheduling latency”) between when a CSI estimate is generated at the UE 110 and when the MIMO management module 320 uses the CSI estimate to control one or more MIMO processes may result in stale CSI estimate information from a previous time period being used to control the MIMO processes for the current time period. To compensate for this scheduling latency, one or more DNNs of the CSF TX processing module 602 may be trained at different scheduling latencies (denoted by the scheduling latency of the BS 108) to provide CSI outputs that are, in effect, predictions of future CSI estimates for the next period, and the predicted future CSI estimates, represented by the CSI outputs 620, are to be used by the MIMO management module 320 of the BS 108. Thus, to this end, the UE 110 may further monitor the scheduling latency of the BS 108. Scheduling latency information 619 representing the time may be provided as an input to the CSF TX processing module 602, which may be determined through analysis of the operation of the BS108, determined based on explicit advertisement of scheduling by the BS108, determined based on the scheduling latency of the BS108 observed by another UE, etc.

[0067] The RF antenna interface 204 and one or more antennas 202 convert this CSI output 620 into a corresponding RF signal 622 that is transmitted over the air for reception by the BS 108. In particular, in some embodiments, one or more DNNs of the CSF TX processing module 602 are trained to provide a process that effectively results in a data-coded (e.g., compressed) representation of the input CSI estimate data block 614, where such a process is trained to the one or more DNNs through joint training rather than requiring tedious and inefficient hard coding of an algorithmic data quantization process. Furthermore, in some embodiments, the one or more DNNs can further operate to effectively provide a channel-coded (including modulation) representation of the CSI estimate data block 614 that is ready for digital-to-analog conversion and RF transmission. That is, rather than using separate discrete processing blocks to implement a data encoding process followed by an initial RF encoding process, the CSF TX processing module 602 can be trained to simultaneously provide the equivalent of such a process, and can be based at least in part on other current data, such as sensor data 616, network status information 618, and scheduling latency information 619, to generate a corresponding signal—in effect, the source is encoded, the channel is encoded, and RF transmission is ready, and modified to reflect predicted future CSI estimates when scheduling latency is taken into account.

[0068] An RF signal 622 propagating from the UE 110 is received and initially processed by the antenna 302 and RF antenna interface 304 of the BS 108. One or more DNNs of the CSF RX processing module 604 are trained to receive the resulting output of the RF antenna interface 304 as inputs 624, optionally along with one or more other inputs, such as sensor data 626 from the sensor set 310 and network status information 628 representing current parameters of the receiving side of the RF antenna interface 304, and generate a corresponding CSI estimate data block 630 that is a recovered representation or version of the CSI estimate data block 614 provided by the CSI estimate module 226 of the UE 110. The processing performed by the CSF RX processing module 604 includes, for example, channel decoding of the input 624 to generate a digital representation of a data-encoded version of the CSI estimate data block 614, as well as effectively decoding (e.g., decompressing) the data itself, generating a decoded representation of the CSI estimate data block 614. The recovered CSI estimate data block 630 may then be provided to the MIMO management module 320 for use in controlling one or more MIMO-based processes between the BS 108 and the UE 110, or between the BS 108 and multiple UEs. As noted above, such operations may include beamforming operations, spatial diversity operations, etc.

[0069] Implementing a jointly trained DNN or other neural network to implement the CSF path between the transmitting and receiving devices provides design flexibility and facilitates efficient updates compared to traditional block-by-block design and testing approaches, but also allows devices in the CSF path to quickly adapt their outgoing and incoming transmit processing to current operating parameters. However, before a DNN can be deployed and begin operation, it is typically trained or otherwise configured to provide appropriate outputs for a specified set of one or more inputs. To this end, FIG. 7 illustrates a typical example of a DNN in some embodiments. 7 illustrates an exemplary method 700 for developing one or more co-trained DNN architecture configurations as an optional selection of devices in a CSF path for different operating environments. It should be noted that the order of operations described with reference to FIG. 7 is for illustrative purposes only, and that operations may be performed in a different order, and further, one or more operations may be omitted, or one or more additional operations may be included in the illustrated method. Furthermore, it should be noted that while FIG. 7 illustrates an offline training approach using one or more test nodes, a similar approach can be implemented for online training using one or more nodes in active operation.

[0070] As described above, the operation of the DNNs used in one or both devices in a DNN chain, forming the corresponding CSF path, may be based on the specific capabilities and current operating parameters of the CSF path, the operating parameters of the device using the corresponding DNN, one or more upstream or downstream devices, or a combination thereof, and the general RF propagation environment. These capabilities and operating parameters may include, for example, the type of sensors used to sense the RF transmission environment of the device, the capabilities of such sensors, the power capacity of one or more devices, the processing capabilities of one or more devices, the RF antenna interface configuration of one or more devices (e.g., number of beams, antenna ports, supported frequencies), etc. Because the described DNNs utilize such information to determine their operation, it will be understood that in many cases the specific DNN configuration implemented in one of the nodes will be based on the specific capabilities and operating parameters currently in use in that device or in the device on the other end of the CSF path. That is, the specific DNN configuration implemented reflects the capability information and operating parameters currently indicated by the CSF paths implemented by the transmitting and receiving devices.

[0071] Thus, method 700 begins at block 702 by identifying the expected capabilities (including expected operating parameters or parameter ranges) of one or more test nodes in a test CSF path, including a test transmitting device and a test receiving device. Hereinafter, it is assumed that the training module 408 of the management component 140 is managing the joint training, and thus the capability information of the test devices is known to the training module 408 (e.g., via a database or other locally stored data structure that stores this information). However, because the management component 140 likely does not have a priori knowledge of the capabilities of a particular UE, the test transmitting device and the test receiving device provide to the management component 140 an indication of their respective capabilities, such as an indication of the types of sensors available on the test devices, various parameters of those sensors (e.g., imaging resolution and image data format for an imaging camera, satellite positioning type and format for a satellite position sensor, etc.), accessories available on the devices and applicable parameters (e.g., number of audio channels), etc. For example, the test equipment may provide an indication of this capability as part of a UECapabilityInformation radio resource control (RRC) message that is typically provided by the UE in response to a UECapabilityEnquiryRRC message sent by the BS, according to at least the 4G LTE and 5G NR specifications. Alternatively, the test UE may provide the indication of the sensor capability as a separate side-channel or control channel communication. Furthermore, in some embodiments, the capabilities of the test equipment may be stored in a local or remote database available to management component 140, so that management component 140 can query this database based on some form of identifier of the test equipment, such as an International Mobile Subscriber Identity (IMSI) value associated with the test equipment.

[0072] In some embodiments, the training module 408 may attempt to train all CSF configuration permutations. However, in implementations where transmitting and receiving devices are likely to have a relatively large number and a wide variety of capabilities and other operating parameters, This approach may be impractical. Therefore, in block 704, the training module 408 may select a specific CSF configuration from a specified set of candidate CSF configurations for jointly training the DNN of the test device. Thus, each candidate CSF configuration may represent a specific combination of CSF-related parameters, parameter ranges, or combinations thereof. Such parameters or parameter ranges may include sensor capability parameters, processing capability parameters, battery power parameters, RF signaling parameters such as the number and type of antennas, the number and type of subchannels, scheduling latency information, etc. Such CSF-related parameters may further represent a specific type of CSI pilot signal used by the transmitting device, a method for the receiving device to calculate CSI estimates, and a format in which the CSI estimates are provided as a CSF, etc. Using the candidate CSF configuration selected for training, further in block 704 the training module 408 identifies an initial DNN architecture configuration for each of the test sending device and the test receiving device and instructs the test device to implement each of these initial DNN architecture configurations by providing the test device with an identifier associated with the initial DNN architecture configuration if the test device stores a copy of the candidate initial DNN architecture configuration or by transmitting data representing the initial DNN architecture configuration itself to the test device.

[0073] Upon selecting a CSF configuration and initializing the test equipment with a DNN architecture configuration based on the selected CSF configuration, in block 706, the training module 408 identifies one or more sets of training data to use in jointly training the DNNs of the DNN chain based on the selected CSF configuration and the initial DNN architecture configuration. That is, the one or more sets of training data include or represent data that can be provided as input to the corresponding DNNs in online operation and thus suitable for training the DNNs. To illustrate, this training data can include streams of test CSI pilot signals, test received representations of the test CSI pilot signals, test sensor data consistent with sensors included in the configuration under test, test network status information consistent with the configuration under test, etc.

[0074] Once one or more training sets have been obtained, the training module 408 begins joint training of the DNNs of the test CSF path at block 708. This joint training typically includes initializing the bias weights and coefficients of the various DNNs with initial values, which are typically selected pseudo-randomly, followed by inputting the set of training data to a TX processing module (e.g., CSF TX processing module 602) of the test receiving device, transmitting the resulting output over the air as a transmission to an RX processing module (e.g., CSF RX processing module 604) of the test receiving device, analyzing the resulting output, and then updating the DNN architecture configuration based on the analysis.

[0075] As is frequently used in DNN training, feedback obtained as a result of the actual resulting output of the CSF RX processing module 604 is used to modify or refine the parameters of one or more DNNs in the CSF path, such as through backpropagation. Thus, at block 710, the management component 140 and / or the DNN chain itself obtains feedback on the training set being transmitted. This feedback can be implemented in any of a variety of forms or combinations of forms. In some embodiments, the feedback includes the training module 408 or other training module determining an error between the actual resulting output and the expected resulting output and backpropagating this error throughout the DNNs in the DNN chain. For example, since processing by the DNN chain effectively provides a form of quantization or other encoding, objective feedback on the training data set may be obtained by modifying the original CSI provided as input to the DNN chain. The accuracy of the recovered CSI estimate data obtained as an output from the DNN chain compared to the estimated value data can be measured in some way. The obtained feedback can also include evaluation metrics regarding some aspects of the signal as it passes through one or more links in the DNN chain. For example, regarding the RF aspects of signal transmission, the feedback can include metrics such as block error rate (BER), signal-to-noise ratio (SNR), signal-to-interference-plus-noise ratio (SINR), etc.

[0076] At block 712, feedback obtained as a result of transmitting the test data set through the DNN chains and presenting or otherwise consuming the resulting output at the test transmitting device is then used to update various aspects of one or more DNNs in the CSF path, for example, through backpropagation of error to modify weights, connections, or layers of the corresponding DNNs, or through managed modifications by management component 140 in response to such feedback. The training process of blocks 706-712 is then performed on the next set of training data selected in the next iteration of block 706, and is repeated until a certain number of training iterations have been performed or until a certain minimum error rate has been achieved.

[0077] As a result of the joint (or individual) training of the neural networks along the CSF paths between the test sending device and the test receiving device, each neural network has a particular neural network architectural configuration, or a DNN architectural configuration if the implemented neural network is a DNN, which characterizes the architecture and parameters of the corresponding DNN, such as the number of hidden layers, the number of nodes in each layer, the connections between each layer, weights, coefficients, and other bias values ​​implemented for each node. Thus, once the joint or individual training of the DNNs of the CSF paths for the selected CSF configurations is completed, at block 714, some or all of the trained DNN configurations are distributed to the BS 108 and UE 110 in the system 100, and each node stores the resulting DNN configuration of the corresponding DNN as a DNN architectural configuration. In at least one embodiment, the DNN architectural configuration can be generated by extracting the architecture and parameters of the corresponding DNN, such as the number of hidden layers, the number of nodes, connections, coefficients, weights, and other bias values ​​at the end of the joint training. In other embodiments, the management component 140 stores copies of the pair of DNN architecture configurations as candidate neural network architecture configurations 144 in the set 142, and these DNN architecture configurations are then distributed to the BSs 108 and UEs 110 as needed.

[0078] If one or more other candidate CSF configurations remain to be trained, then method 700 returns to block 704 to select the next candidate CSF configuration to be jointly trained, and another iteration of the subprocess of blocks 704-714 is repeated for the next CSF configuration selected by training module 408. Otherwise, if the CSF path DNNs have been jointly trained for all intended CSF configurations, method 700 is complete, and system 100 can move on to neural network-supported CSI estimate feedback, as described below with reference to FIGS.

[0079] As noted above, the joint training process can be performed using offline test nodes (i.e., while no active communication of control information or user plane data is occurring) or while the actual nodes in the intended transmission path are online (i.e., while active communication of control information or user plane data is occurring). Furthermore, in some embodiments, rather than training all DNNs together, a subset of DNNs may be trained or retrained, possibly while other DNNs remain static. To illustrate, a management component 140 may detect that a DNN of a particular device is operating inefficiently or incorrectly, for example, due to the presence of an undetected source of interference near the device implementing the DNN or in response to a previously unreported loss of processing power, and thus, management component 140 may schedule an individual retraining of the device's DNN while maintaining other DNNs of other devices in their current configurations.

[0080] Furthermore, it will be understood that there may be a wide variety of devices supporting numerous CSF configurations, and that many different nodes may support the same or similar CSF configurations. Thus, following joint training of a representative device, rather than having to repeat joint training of all devices to be incorporated into a CSF path, the device may transmit a representation of the DNN architecture configuration to be trained for the CSF configuration to management component 140, which may store the DNN architecture configuration and subsequently transmit it to other devices supporting the same or similar CSF configuration for implementation in the DNN of the CSF path.

[0081] Furthermore, DNN architecture configurations often change over time as the corresponding device operates using the DNN. Thus, as operation progresses, a particular device's neural network management module (e.g., neural network management module 222, 314) may be configured to transmit a representation of an updated architecture configuration of one or more DNNs used at that node. For example, by providing updated gradients and related information to management component 140 in response to a trigger. This trigger may be the expiration of a periodic timer, a query from management component 140, a determination that the magnitude of change exceeds a specified threshold, etc. Management component 140 then incorporates these received DNN updates into the corresponding DNN architecture configuration, thus having an updated DNN architecture configuration that can be distributed to nodes in the transmission path as needed.

[0082] 8 and 9 jointly illustrate an example method 800 for channel state feedback using jointly trained DNN-based CSF paths between wireless devices, in accordance with some embodiments. For ease of explanation, the method 800 of FIG. 8 is described below with reference to the example CSF paths 116 of FIGS. 1 and 6, with the BS 108 acting as the transmitting device and the UE 110 acting as the receiving device. Additionally, the process of the method 800 is described with reference to the example transaction (ladder) diagram 900 of FIG. 9.

[0083] The method 800 begins at block 802, where the BS 108 and the UE 110 establish a wireless connection, such as via a 5G NR standalone registration / attach process in a cellular context or an IEEE 802.11-related process in a WLAN context. At block 804, the management component 140 obtains capability information from each of the BS 108 and the UE 110, such as capability information 902 (FIG. 9) provided by the capability management module 316 (FIG. 3) of the BS 108 or capability information 904 (FIG. 9) provided by the capability management module 224 (FIG. 2) of the UE 110. In some embodiments, the management component 140 may already be informed of the capabilities of the BS 108 if the BS 108 is part of the same infrastructure network. In this case, obtaining the capability information 902 of the BS 108 may include accessing a local or remote database or other data store for this information. In the case of the UE 110, the BS 108 may send a capability request to the UE 110, to which the UE 110 responds with capability information 904, which the BS 108 then forwards to the management component 140. For example, the BS 108 may send a UECapabilityEnquiry RRC message, to which the UE 110 responds with a UECapabilityInformation RRC message containing CSI-related capability information.

[0084] At block 806, the neural network selection module 410 of the management component 140 uses the capability information and other information describing the CSF configuration between the BS 108 and the UE 110 to select a pair of CSF DNN architecture configurations to be implemented at the BS 108 and the UE 110 to support the CSF path 116 (DNN selection 906, FIG. 9 ). In some embodiments, the neural network selection module 410 uses an algorithmic selection process in which the capability information obtained from the BS 108 and the UE 110 and the CSF configuration parameters of the CSF path 116 are compared with attributes of a pair of candidate neural network architecture configurations 144 in the set 142 to identify a suitable pair of DNN architecture configurations. In other embodiments, the neural network selection module 410 can organize the candidate DNN architecture configurations into one or more LUTs, with each entry storing a corresponding pair of DNN architecture configurations and indexed by a corresponding combination of input parameters or parameter ranges, and thus the neural network selection module 410 can select an appropriate pair of DNN architecture configurations to be used by the BS 108 and the UE 110 via providing the capability and CSF configuration parameters identified in block 804 as inputs to one or more LUTs.

[0085] Further, at block 806, the management component 140 instructs the BS 108 and the UE 110 to implement respective DNN architecture configurations from the pair of selected co-trained DNN architecture configurations. In implementations in which the BS 108 and the UE 110 each store candidate DNN architecture configurations for potential future use, the management component 140 can transmit a message including an identifier of the DNN architecture configuration to be implemented. Otherwise, the management component 140 can transmit information representing the DNN architecture configuration, for example, as a layer 1 signal, a layer 2 control element, a layer 3 RRC message, or a combination thereof. For example, referring to FIG. 9 , the management component 140 transmits a DNN configuration message 908 to the BS 108 including data representing the DNN architecture configuration selected for the BS 108. In response to receiving this message, the neural network management module 314 of the BS 108 extracts data from the DNN configuration message 908 and configures the CSF RX processing module 604 to implement one or more DNNs having the DNN architecture configuration represented by the extracted data. Similarly, the management component 140 transmits a DNN configuration message 910 to the UE 110 that includes data representing a DNN architecture configuration selected for the UE 110. In response to receiving this message, the neural network management module 222 of the UE 110 extracts the data from the DNN configuration message 910 and configures the CSF TX processing module 602 to implement one or more DNNs having the DNN architecture configuration represented by the extracted data.

[0086] Once the DNN for the CSF path 116 is initially configured, the CSI estimation and feedback process can begin. Thus, in block 808, the CSF management module 318 of the BS 108 selects or identifies a CSI pilot signal 912 (FIG. 9) based on the CSF configuration for the CSF path 116 (which may include, for example, the particular beam, antenna, subcarrier, etc. used) and provides wireless transmission of the CSI pilot signal 912 to the UE 110. In block 810, the UE 110 receives one or more RF signals representing the CSI pilot signal 912 to be transmitted and converts the one or more RF signals to one or more corresponding baseband signals; the CSI estimation module 226 then analyzes the one or more baseband signals to determine a CSI estimate 914 (FIG. 9). As described above, the CSI estimation module 226 can determine a CSI estimate using any of a variety of CSI estimation techniques.

[0087] At block 812, the CSF TX processing module 602 receives as input the CSI estimate 914, optionally along with one or more other inputs, such as sensor data from a sensor in the UE 110, scheduling latency information representing the scheduling latency in using the CSI estimate at the BS 108, and / or current network status information from the RF antenna interface 204 of the UE 110, and generates from these inputs a CSF output 916 ( FIG. 9 ) representing the CSI estimate 914 in a quantized or coded form. At block 814, the resulting CSF output 916 is transmitted wirelessly from the UE 110 to the BS 108.

[0088] At block 816, one or more RF signals representing the CSF output 916 are received and processed by the RF front end 304 of the BS 108, and the resulting output, optionally along with one or more other inputs such as sensor data from the sensor set 310 of the BS 108, current network status information obtained by the BS 108, etc., is used to generate the CSF of the BS 108. The CSF TX processing module 602 of the UE 110 may further consider the scheduling latency of the BS 108 when generating the CSI output, and thus the CSI estimates 914 are modified during processing by the CSF TX processing module 602 to reflect predicted future versions of the CSI estimates. Thus, the recovered CSI estimates 918 represent predicted CSI estimates during the time periods during which the MIMO management module 320 is scheduled to use the CSI estimates in controlling the MIMO processes.

[0089] Typically, the CSI estimation process involves transmitting a sequence of CSI pilot signals, each CSI pilot signal (or a subset of CSI pilot signals) configured for use in characterizing a particular subchannel or carrier frequency among a set of subchannels / carrier frequencies. Thus, the process of blocks 808 through 818 can be repeated for each CSI pilot signal in such a sequence. For example, in the next iteration of this process, a CSI pilot signal 922 (FIG. 9) can be selected for channel estimation of another subcarrier and transmitted from the BS 108 to the UE 110. The CSI estimation module 225 processes a received version of the CSI pilot signal 922 to determine a CSI estimate 924 (FIG. 9) for that subcarrier, which is provided as an input, along with other inputs, to the CSF TX processing module 602, which generates a coded representation of the CSI estimate 924 in the form of a CSF output 926 (FIG. 9). The CSF output 926 is then transmitted wirelessly to the BS 108, whereupon a recovered representation of the CSF output 926 is 9. The CSI estimate 924 is provided as an input to the RX processing module 604, which uses this input and optionally one or more other inputs to generate a reconstructed representation of the CSI estimate 924 (reconstructed CSI estimate 928 in FIG. 9), which can then be used to modify or control one or more MIMO processes 930 (FIG. 9) at the BS 108.

[0090] 8 and the corresponding operational example of ladder diagram 900 of FIG. 9 illustrate an implementation in which each CSI estimate generated at UE 110 is used to generate a corresponding individual CSF output. In other embodiments, rather than generating a new CSF output for each CSI estimate generated at UE 110, UE 110 may be configured to generate and temporarily store CSI estimates for some or all of the entire set of CSI pilot signals transmitted by BS 108 for a particular CSF iteration, and then combine the resulting set of CSI estimates into a single data block (e.g., in the form of a CSI estimate matrix or other data structure) (e.g., as shown in FIG. 9). 6. The CSI estimates are provided as input to the CSF TX processing module 602, which generates a single CSF output representing the set of CSI estimates, as a single CSI estimate data block 134.

[0091] Thus far, system 100 has been described in connection with an embodiment in which a conventional CSI pilot transmission and CSI estimation process is used in conjunction with a neural network-based channel state feedback path to provide the resulting CSI estimates to the transmitting device. However, system 100 is not limited to this approach and may instead use a DNN or other neural network jointly trained for each of the CSI pilot transmission, CSI estimation, and CSI feedback processes. To this end, FIG. 10 illustrates an example operating environment 1000 in which the CSI pilot transmission, CSI estimation, and CSI feedback processes are implemented using a set of DNNs or other neural networks jointly trained at the transmitting device and the receiving device. For ease of explanation, the operating environment is described in the example context of a BS 108 as a transmitting device and a UE 110 as a receiving device. The disclosed principles apply equally to examples with the UE 110 as a transmitting device and the BS 108 as a receiving device, as well as to sidelink examples.

[0092] In the illustrated example operating environment 1000, the neural network management module 314 of the BS 108 employs a CSF TX processing module 1002 and a CSF RX processing module 1008, while the neural network management module 222 of the UE 110 implements a CSF RX processing module 1004 and a CSF TX processing module 1006. In at least one embodiment, each of these processing modules implements one or more DNNs via implementation of a corresponding ML module, as described above with reference to one or more DNNs 502 of the ML module 500 of FIG.

[0093] Processing modules 1002, 1004, 1006, and 1008 work together to provide CSFs in a manner that reduces or eliminates the use of complex hard-coded CSF processes in favor of a trained neural network-based approach that can adapt to current operating parameters of BS 108 and UE 110, as well as changes in these operating parameters. To this end, candidate neural network architecture configurations can be jointly trained as described above with reference to method 700 of FIG. 7, and management component 140 or other component of system 100 can select a particular set of neural network architecture configurations to be used for these processing modules from the jointly trained candidates using a selection process similar to that described above.

[0094] FIG. 11 illustrates an example of a method 1100 of operation of the operating environment 1000 of FIG. 10 in accordance with some embodiments. For ease of understanding, the method 1100 will be described with reference to the example transactional (ladder) diagram 1200 of FIG. 12. The method 1100 begins at block 1102, where the BS 108 and the UE 110 establish an initial connection, as described above with reference to blocks 802 and 804, and the management component 140 obtains CSF-related capability information from the BS 108 and the UE 110 (FIG. 8). As also described above with reference to block 806, the management component 140 selects a set of DNN architecture configurations to be used by the processing modules 1002, 1004, 1006, and 1008 based on the obtained capability information and any CSF configuration information provided, and instructs the BS 108 and the UE 110 to implement the selected DNN architecture configurations. This process is represented in the transactional diagram 1200 as a DNN configuration process 1202.

[0095] At block 1104, the CSF management module 318 generates a CSF configuration input 1010 (FIGS. 10 and 12) comprising information representing the current CSF configuration, and communicates the CSF configuration to the BS 108 and the UE 11. 0 will collaborate to provide CSI for use by BS 108 in downlink communications with UE 110. This information may include, for example, the particular subset of beams to be used, the particular subset of carrier frequencies to be used, the particular number of antenna ports to be used, a specified power efficiency target, an indication of expected RF signaling and whether it is line of sight (LoS), multipath, or a combination thereof.

[0096] At block 1106, the CSF configuration input 1010 is provided as an input to the CSF TX processing module 1002 of the BS 108, optionally along with one or more other inputs, such as current sensor data from the sensor set 310 or current network status information observed by the BS 108 (for ease of illustration, these inputs are omitted from FIG. 10 ). Based on the CSF configuration input 1010 and the other inputs, the CSF TX processing module 1002 operates to generate a CSI pilot output 1012 ( FIGS. 10 , 12 ), which effectively represents one or more CSI pilot signals, reflecting the current CSF configuration represented by the CSF configuration input 1010. The CSI pilot output 1012 is then transmitted wirelessly to the UE 110 at block 1108. Additionally, the CSI pilot output 1012 is also provided as an input to the CSF RX processing module 1008 of the BS 108, which optionally uses this input along with other inputs such as sensor data and current network status information to recover CSI estimate information from the DNN-generated output at the UE 110, as described below.

[0097] At block 1110, the UE 110 processes the received wireless signal representing the CSI pilot output 1012 and provides the result, optionally along with one or more other inputs such as sensor data from the UE 110's sensor set 210 or current network status information observed by the UE 110, as input to the UE 110's CSF RX processing module 1006 (omitted from FIG. 10 for ease of illustration). The RX processing module 1006 uses these inputs to generate corresponding CSI outputs 1014 (FIGS. 10, 12), which effectively represent one or more CSI estimates for some or all of the subchannel / beam / antenna port combinations represented by the CSF configuration information 1010 and generated from the CSI pilot information represented by the received CSI pilot outputs 1012. That is, the CSF TX processing module 1002 and the CSF RX processing module 1006 can be jointly trained to effectively provide the equivalent of a conventional algorithmic CSI pilot transmission and CSI estimate calculation process, but the processing modules 1002 and 1006 can further incorporate current sensor inputs and current network conditions to better adapt current operating parameters to the current transmission environment, as well as scheduling latency information to better predict the CSI estimate information for the period during which the BS 108 will use the CSI estimate information being generated.

[0098] At block 1112, the CSF TX module 1006 receives as input the CSI output 1014, optionally along with sensor data, network status data, or other data as one or more other inputs, from which, in effect, a CSF output 1016 (FIGS. 10, 12) is generated, representing a compressed or encoded representation of the CSI estimate information represented by the CSI output 1014, adapted to the current operating context represented by the sensor data, network status information, scheduling latency, and other inputs. At block 1114, the UE 110 wirelessly transmits the CSF output 1016 to the BS 108.

[0099] In block 1116, the BS 108 processes the wireless signal representing the CSF output 1016 and provides a resulting output representing a recovered CSI representation of the CSF output 1016 to the CSF RX processing module 1008. The CSF RX processing module 1008 converts this input into , along with the CSI pilot output 1012 as an input, and optionally further along with one or more other inputs, such as sensor data observed at the BS 108 and current network status information, to generate a CSI estimate output 1018, which represents a recovered representation of the CSI estimate information represented in the CSI output 1014 generated by the CSF RX processing module 1004 of the UE 110 based on a received representation of the CSI pilot output 1012 generated by the CSF TX processing module 1002 of the BS 108. That is, in one embodiment, the DNNs of processing modules 1006 and 1008 are jointly trained to effectively provide the equivalent of a traditional hard-coded CSI feedback process, where the CSI estimates are coded for transmission, but processing modules 1006 and 1008 can further incorporate current operating parameters in the form of current sensor inputs and current network conditions to better adapt to the current transmission environment. At block 1118 , the CSI estimate information, represented by the CSI estimate output 1018 , is then provided to the MIMO management module 320 for use in controlling one or more MIMO processes being used at the BS 108 .

[0100] Typically, changes in the operating environment may require recalibration or recalculation of the CSI estimates provided during the most recent iteration of the process of blocks 1104-1118. For example, the location of the UE 110 may change sufficiently to require recalculation, the antenna patterns of one or both of the BS 108 or the UE 110 may have changed substantially, etc. Accordingly, the process of blocks 1104-1118 may be repeated to update the CSI estimates used by the BS 108 to control certain MIMO operations. Triggering another iteration of the process of blocks 1104-1118 may be based on a timer or other periodic criteria. For example, the BS 108 or the management component 140 may trigger another iteration based on the expiration of a timer. In other embodiments, the timing of the iteration trigger may be trained into the DNN itself, thereby allowing the CSI of the BS 108 to be updated. The TX processing module 1002 may be trained to trigger another iteration based on a timer, a particular sensor data input or network status input, or the like.

[0101] In some embodiments, certain aspects of the above-described techniques may be implemented by one or more processors of a processing system executing software. The software includes one or more sets of executable instructions stored or otherwise tangibly embodied on a non-transitory computer-readable storage medium. The software may include instructions and specific data that, when executed by one or more processors, operate the one or more processors to perform one or more aspects of the above-described techniques. Non-transitory computer-readable storage media may include, for example, magnetic or optical disk storage, solid-state storage devices such as flash memory, cache, random access memory (RAM), or other non-volatile memory devices. The executable instructions stored on the non-transitory computer-readable storage medium may be source code, assembly language code, object code, or another instruction format interpretable or executable by one or more processors.

[0102] A computer-readable storage medium may include any storage medium, or combination of storage media, that accesses a computer system during use to provide instructions and / or data to the computer system. Such storage media may include, but are not limited to, optical media (e.g., compact discs (CDs), digital versatile discs (DVDs), Blu-ray discs), magnetic media (e.g., floppy disks, magnetic tape, magnetic hard drives), volatile memory (e.g., random access memory (RAM) or cache), non-volatile memory (e.g., read-only memory (ROM) or flash memory), or microelectromechanical systems (MEMS)-based storage media. A computer-readable storage medium may be embedded in a computer system (e.g., system RAM or ROM). The storage device may be permanently connected to the computing system (e.g., a magnetic hard drive), removably attached to the computing system (e.g., an optical disk or Universal Serial Bus (USB)-based flash memory), or connected to the computer system via a wired or wireless network (e.g., network-accessed storage (NAS)).

[0103] It should be noted that not all of the activities or elements described above in the general description are required, that some of the specific activities or devices may not be required, and that one or more additional activities may be performed or elements may be included in addition to those described. Furthermore, the order in which the activities are listed is not necessarily the order in which they are performed. Also, the concepts are described with reference to specific embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of the present disclosure, as set forth in the claims below. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense, and all such modifications are intended to be included within the scope of the present disclosure.

[0104] Advantages, other advantages, and solutions to problems have been described above with regard to specific embodiments. However, advantages, benefits, solutions to problems, and features from which the advantages, benefits, solutions arise or become more pronounced should not be construed as critical, necessary, or essential features of any or all of the claims. Moreover, the specific embodiments disclosed above are merely exemplary, as the disclosed subject matter may be modified and practiced in different but equivalent manners apparent to those skilled in the art having the benefit of the teachings herein. No limitations are intended to the details of construction or design herein shown, except as set forth in the claims below. It will therefore be apparent that the specific embodiments disclosed above may be altered or modified, and all such variations are considered to be within the scope of the disclosed subject matter. Accordingly, the protection sought herein is as set forth in the following claims.

Claims

1. A computer-implemented method on a first device, comprising: receiving an indication of a neural network architecture configuration in response to providing capability information representing at least one capability of the first device to an infrastructure component; implementing the neural network architecture configuration in a transmitting neural network of the first device; receiving a representation of a channel state information (CSI) estimate as an input to the transmit neural network; generating, in the transmit neural network, a first output based on the representation of the CSI estimate, the first output representing a condensed version of the representation of a prediction of the CSI estimate for a future time point; and A computer-implemented method comprising controlling a radio frequency (RF) antenna interface of the first device to transmit a first RF signal representative of the first power output for reception by a second device.

2. 10. The method of claim 1, The method further comprising algorithmically determining the CSI estimate based on one or more RF signals received from the second device.

3. 2. The method of claim 1 , wherein generating the first output further comprises generating the first output with the transmit neural network further based on a representation of a scheduling latency of a multiple-input multiple-output (MIMO) process of the second device provided as an input to the transmit neural network.

4. The method of claim 3 , wherein the neural network architectural configuration is selected for the transmit neural network from a plurality of candidate neural network architectural configurations based on the scheduling latency.

5. 5. The method of claim 1, wherein generating the first output comprises generating the first output in the transmit neural network further based on sensor data input to the transmit neural network from one or more sensors of the first device.

6. receiving a representation of a CSI pilot signal as an input to a receiver neural network of the first device; and and generating, in the receiving neural network, a second output based on the representation of the CSI pilot signal, the second output comprising the representation of the CSI estimate.

7. 7. The method of claim 6, wherein generating the second output further comprises generating the second output with the receiver neural network based on at least one of sensor data from one or more sensors of the first device or a carrier frequency of a channel associated with the CSI estimate.

8. A computer-implemented method on a first device, comprising: receiving an indication of a neural network architecture configuration in response to providing capability information representing at least one capability of the first device to an infrastructure component; implementing the neural network architecture configuration in a receiving neural network of the first device; receiving, at an RF antenna interface of the first device, a first RF signal representing a compressed representation of a predicted future channel state information (CSI) estimate from a second device; providing a representation of the first RF signal as an input to the receiver neural network; generating, at the receiver neural network, the predicted future CSI estimate based on the input to the receiver neural network; and and managing at least one multiple-input multiple-output (MIMO) process at the first device based on the predicted future CSI estimate.

9. 9. The method of claim 8, wherein generating the predicted future CSI estimates further comprises generating the predicted future CSI estimates with the receiver neural network further based on a representation of a scheduling latency of a multiple-input multiple-output (MIMO) process of the first device provided as an input to the receiver neural network.

10. 10. The method of claim 9, wherein the neural network architectural configuration is selected for the receiving neural network from a plurality of candidate neural network architectural configurations based on the scheduling latency.

11. The method of any one of claims 8 to 10, wherein generating the predicted future CSI estimate comprises generating the predicted future CSI estimate in the receiving neural network further based on sensor data input to the receiving neural network from one or more sensors of the first device.

12. 11. The method of claim 1, wherein the neural network architecture configuration is selected from a plurality of neural network architecture configurations based on at least one of the at least one capability of the first device or a current signal propagation environment of the first device.

13. The step of receiving the indication of the neural network architecture configuration comprises: receiving an identifier associated with one of a plurality of candidate neural network architecture configurations stored locally on the first device; or and receiving one or more data structures representing parameters of said neural network architecture configuration.

14. generating a CSI pilot signal with a transmit neural network of the first device; and 14. The method of claim 8, further comprising controlling the RF antenna interface of the first device to transmit a second RF signal representing the CSI pilot signal for reception by the second device.

15. The step of generating the CSI pilot signal includes generating the CSI pilot signal in the transmit neural network further based on at least one of a carrier frequency of a channel associated with the CSI estimate, at least one operating parameter of the RF antenna interface of the first device, sensor data from one or more sensors of the first device, or a carrier frequency of a channel associated with the CSI estimate.

15. The method of claim 14, further comprising generating a code.

16. The method of any one of claims 8 to 15, wherein generating the predicted future CSI estimate further comprises generating, in the transmit neural network, the predicted future CSI estimate based on at least one of sensor data from one or more sensors of the first device or a carrier frequency of a channel associated with the predicted future CSI estimate.

17. The method according to any one of claims 8 to 16, wherein the at least one MIMO process includes at least one of a beamforming process, a space-time coding process, or a multi-user MIMO process.

18. The method of any preceding claim, wherein the at least one capability includes at least one of processing capability, power capability, or sensor capability.

19. a radio frequency (RF) antenna interface; at least one processor coupled to the RF antenna interface; and a memory storing executable instructions, the executable instructions configured to operate the at least one processor to perform a method according to any one of claims 1 to 18.

Images